Automaton Instruction (AI)
Jonathan Schaeffer · 2026
In this chapter, the two most important types of machine learning approaches are discussed: supervised learning and reinforcement learning. Today’s popular supervised learning method is called a neural network (NN). The name is an analogy as to how the brain processes information and learns. The NN has inputs and outputs. In between are hidden layers each of which takes input from the previous layer, refines it, and passes it to the next layer. An NN is fed, say, a face and is asked to name the person. A small reward is given if it names the right person, and a small penalty if it doesn’t. This is percolated back through the NN, rewarding or penalizing the parts of the NN responsible for the decision. Thus, each training example helps the NN learn to give the right answer. Such a network was able to learn to identify objects at a superhuman level. Reinforcement learning is experience based. The learning agent tries something (e.g., driving) and eventually gets rewarded (positive or negative). It feeds the reward back through its learning process, making adjustments just like in a NN. The difference is that the RL agent learns from experience, not from a large, labeled data set.